Drip irrigation pattern evaluation and optimization method and system based on fusion ecological model

Abstract

The present application relates to the technical field of intelligent management and control of agricultural irrigation, and particularly relates to a drip irrigation mode evaluation and optimization method and system fusing an ecological model, and specifically as follows: first, collecting multi-source data of an irrigation plot and completing standardization processing; through dynamic time alignment and adaptive kernel mapping with soil salinity constraints, a dynamic collaborative feature matrix retaining coupling relationship is constructed; a particle swarm optimizer fusing ecological model constraints is constructed to perform multi-objective optimization on drip irrigation mode parameters to obtain a Pareto optimal solution set; a comprehensive evaluation system containing yield marginal benefit and uncertainty penalty is established to screen an optimal scheme; finally, through field correction and rolling optimization, closed-loop management and control is realized. The present application can solve the problems of data coupling relationship damage, ecological risk constraint absence and optimization scheme robustness deficiency in existing drip irrigation optimization methods, realize multi-objective optimization of drip irrigation mode, take into account high yield benefit and ecological safety, and improve scheme robustness and field applicability.

Description

Technical Field

[0001] This invention relates to the field of intelligent management and control technology for agricultural irrigation, and in particular to a method and system for evaluating and optimizing drip irrigation patterns that integrates an ecological model. Background Technology

[0002] With the increasingly severe shortage of agricultural water resources, drip irrigation technology has been widely used globally due to its high efficiency and water-saving characteristics. However, in actual production, the management decisions of drip irrigation systems often face multiple challenges. On the one hand, drip irrigation operations involve multiple decision variables, such as irrigation volume, irrigation interval, and the ratio and application rate of nitrogen, phosphorus, and potassium fertilizers. These variables have complex coupling relationships, and there is a significant time lag between crop growth response and ecological environment feedback relative to irrigation behavior, making it difficult to achieve global optimization using traditional experience-based decision-making or single-objective optimization. On the other hand, drip irrigation systems have gradually exposed long-term ecological risks during their promotion and application, especially in arid and semi-arid regions. Inappropriate drip irrigation fertilization patterns can easily lead to soil salt accumulation in the root zone and aggravated nitrate nitrogen leaching, which in turn can cause secondary soil salinization, groundwater pollution, and other ecological problems, thus restricting the sustainable development of drip irrigation technology.

[0003] Existing technologies objectively suffer from the following shortcomings: Most current drip irrigation optimization methods simply interpolate or fill in the mean values ​​of irrigation parameters, crop parameters, and ecological indicators, then directly concatenate them. This fails to address issues of inconsistent sampling frequencies and response lags, disrupting the causal coupling between irrigation behavior, crop response, and ecological feedback. Consequently, optimization models struggle to identify the true correlation between high yield and low ecological risk. Conventional particle swarm optimization or genetic algorithms rely solely on single numerical objectives such as yield or water use efficiency in drip irrigation parameter optimization, lacking dynamic constraints on ecological processes like soil salinity accumulation and nitrate nitrogen leaching. This can easily converge to mathematically optimal methods that result in continuously increasing ecological risks during field application. Unsustainable solutions: Existing methods often employ fixed weighted summation or simple Pareto ranking in multi-objective decision-making, ignoring the diminishing marginal returns of yield improvement in actual agricultural decision-making. Furthermore, they fail to consider the impact of uncertainties in model parameters and field environment on the stability of recommended solutions, resulting in insufficient robustness of output solutions in practical applications. Most drip irrigation optimization research remains at the level of offline simulation or theoretical recommendation, lacking a closed-loop linkage mechanism with field monitoring equipment and intelligent control platforms. It is impossible to use validation experimental data to calibrate the model, nor can it dynamically adjust subsequent irrigation strategies based on the real-time status of crops during their growth period. As a result, optimization results are difficult to implement and continuously improve in actual systems.

[0004] Therefore, this invention proposes a method and system for evaluating and optimizing drip irrigation patterns by integrating ecological models to solve the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a method and system for evaluating and optimizing drip irrigation patterns that integrates ecological models. This invention can achieve multi-objective optimization of drip irrigation patterns, taking into account both high yield and ecological safety, and improving the robustness and field applicability of the scheme.

[0006] On the one hand, the technical solution of this invention to solve the technical problem is a drip irrigation mode evaluation and optimization method integrating an ecological model, including:

[0007] Collect multi-source data from irrigated plots and perform standardized processing;

[0008] The standardized data is dynamically aligned temporally and an adaptive kernel function is constructed to build a dynamic collaborative feature matrix.

[0009] Introducing ecological model constraints, adaptive inertia weights, and ecological correction terms into the particle swarm optimization framework, a particle swarm optimizer integrating ecological model constraints is constructed. Combining the dynamic collaborative feature matrix and crop growth model simulation results, the search direction of the particle swarm optimizer is normally corrected.

[0010] After particle swarm optimization, Pareto optimal solution is obtained, a comprehensive evaluation system is constructed, and the optimal drip irrigation mode is selected from the Pareto optimal solution;

[0011] The optimal drip irrigation mode is evaluated and optimized. After field verification and correction, the optimal drip irrigation mode is integrated into the intelligent control platform. A rolling optimization mechanism is introduced for dynamic adaptive regulation and supports customized solution output. Finally, a drip irrigation mode evaluation and optimization application report is generated.

[0012] The data collection process is as follows:

[0013] Collect multi-source raw data covering multiple irrigation cycles or the entire growth stage, and systematically label the collected data samples;

[0014] The multi-source raw data includes irrigation parameters from irrigation control equipment in the drip irrigation system, crop growth parameters and ecological indicators from field monitoring equipment; irrigation parameters include irrigation volume per irrigation, irrigation interval, nitrogen fertilizer application rate, phosphorus fertilizer application rate, and potassium fertilizer application rate; crop growth parameters include crop plant height, stem diameter, leaf area index, aboveground biomass, leaf chlorophyll content, and canopy temperature; ecological indicators include soil salinity, soil nitrate nitrogen content, soil moisture content, and groundwater depth; all observation records for each plot within a complete irrigation cycle or a complete growth stage are defined as a sample.

[0015] Each sample was labeled with irrigation mode, crop response, and ecological feedback tags.

[0016] A unified time index is established for the samples, and the entire monitoring period is divided into multiple consecutive reproductive stages or standardized time points. The collected multi-source data are associated with the time index according to their actual observation time points to form a labeled dataset.

[0017] Constructing a dynamic collaborative feature matrix:

[0018] The collected multi-source raw data are divided into raw irrigation parameter matrix, raw crop growth parameter matrix and raw ecological index matrix according to data type. Each matrix is ​​standardized by column, and the standardization results are checked for outlier consistency. The standardized irrigation parameter matrix, standardized crop growth parameter matrix and standardized ecological index matrix are output.

[0019] Based on the growth patterns of target crops on irrigated plots, the entire growth period is divided into a unified stage sequence, resulting in a unified growth stage time axis. Based on this unified growth stage time axis, standardized multi-source data is reorganized into sample-level time-series vectors. Dynamic time warping is then performed on multiple types of time-series vectors within the same sample. Following the alignment path obtained from the dynamic time warping, the original unequal-length sequences are mapped onto the unified growth stage time axis. Soil salinity time-series data is extracted from this mapping to construct a soil salinity constraint sequence, where soil salinity data represents ecological indicator parameters from the multi-source data. Sample-level time-series vectors mapped onto the unified growth stage time axis are defined to have equal lengths and a one-to-one correspondence between stage points. The soil salinity constraint sequence is introduced, and an adaptive kernel function is constructed using the aligned sample-level time-series vectors as input.

[0020] The aligned sample-level time-series vectors are concatenated in a fixed order. For any two samples, the kernel value between the two samples is calculated using an adaptive kernel function to obtain an adaptive kernel matrix. The adaptive kernel matrix is ​​then centered, and eigenvalue decomposition is performed to obtain the kernel principal component directions sorted by contribution. The first few principal components are retained according to a preset rule, and each sample is projected onto the retained kernel principal component directions to obtain a dynamic collaborative feature matrix.

[0021] The process of building a particle optimizer is as follows:

[0022] The particle swarm optimizer is initialized, and the drip irrigation mode parameter vector is defined as the encoding method of the particles. The corresponding state vector in the dynamic collaborative feature matrix and the drip irrigation mode parameter vector of each particle are input into the model evaluation module, which includes the soil salinity balance model, crop growth model and nitrogen balance model. Based on the initial evaluation results of the model evaluation module, the individual historical optimal position of each particle is set, and an external non-dominated solution archive is established.

[0023] Calculate the base velocity term and the ecological correction term, add the two together to get the new particle velocity, and set an upper limit for the velocity in each dimension;

[0024] The particle position is iteratively updated based on the updated particle velocity. The feasible region is corrected for components that exceed the upper and lower limits of the parameters. A combination check is performed on the agronomic constraints that characterize the relationship between irrigation and fertilizer ratio. If the new position does not meet the combination constraints, it is projected to the feasible solution closest to the current point. The model evaluation module is called again. The historical best position of the individual particle is updated according to the Pareto dominance relationship. At the same time, the external archive is updated, the dominated solutions are deleted, and the new non-dominated solutions are retained.

[0025] Repeatedly perform particle velocity updates, particle position updates, and archive updates. After multiple iterations, the particle swarm optimization is completed.

[0026] The calculation process for the basic velocity term and the ecological correction term is as follows:

[0027] Based on the external non-dominated solution archive, a global guidance position is selected for each particle. The inertial weight is dynamically corrected based on the soil salinity component corresponding to the global guidance position, and the adaptive inertial weight is calculated.

[0028] The basic velocity term is calculated based on the standard particle swarm update mechanism. This term includes the inertial inheritance term of the previous generation velocity determined by the adaptive inertial weight, the individual learning term toward the individual's historical best position, and the group learning term toward the global guiding position.

[0029] The ecological costs at each reproductive stage are obtained through forward simulation. The ecological sensitivity vector is calculated by finite difference perturbation of decision variables. The correction term along the direction of ecological risk reduction is generated by combining the yield status adjustment correction intensity.

[0030] The process of obtaining the Pareto optimal solution is as follows:

[0031] After particle swarm optimization is completed, all non-dominated solutions are extracted from the external non-dominated solution archive to form a Pareto optimal solution set. The Pareto optimal solution set is then deduplicated and merged with nearest neighbors.

[0032] Pareto optimal solution set includes If two candidate drip irrigation modes have different parameters in each dimension, and the difference between them is less than the set tolerance, then the relatively stable candidate drip irrigation mode is retained.

[0033] The process of the comprehensive evaluation system is as follows:

[0034] Based on the forward simulation results of each candidate drip irrigation mode using crop growth model, soil salinity balance model and nitrogen balance model, the yield benefits and ecological benefits of each candidate drip irrigation mode in the Pareto optimal solution set are calculated.

[0035] The yield and ecological benefits of all candidate drip irrigation models in the Pareto optimal solution set were statistically analyzed to obtain the minimum, maximum, minimum, and maximum yield benefits, respectively. Uncertainty analysis was performed on each candidate drip irrigation model to obtain the standard deviation of ecological risk.

[0036] For each candidate drip irrigation mode, a nonlinear utility mapping is performed on the yield benefit index to obtain the yield utility value; for each candidate drip irrigation mode, the ecological benefit index is normalized to obtain the ecological utility value.

[0037] The yield utility value, ecological utility value, and ecological risk penalty are combined into a comprehensive score. The candidate drip irrigation mode with the highest comprehensive score is selected as the final recommended scheme from all candidate drip irrigation modes.

[0038] The calculation process for yield benefits and ecological benefits is as follows:

[0039] Pareto optimal solution set includes For each candidate drip irrigation mode, the crop growth model, soil salinity balance model, and nitrogen balance model are called for forward simulation. At the same time, the parameter vector of the candidate drip irrigation mode and the dynamic collaborative feature vector of the corresponding scenario are input. Based on the crop growth model, the yield benefit index of the candidate drip irrigation mode is output.

[0040] The output of the soil salinity balance model is as follows: Soil salt accumulation for each candidate drip irrigation pattern, output according to the nitrogen balance model. The nitrate nitrogen leaching loss of each candidate drip irrigation mode is calculated, and the irrigation water use efficiency of the candidate drip irrigation mode is calculated based on the total yield and total irrigation water volume. Soil salinity accumulation and nitrate nitrogen leaching loss are converted into attenuation terms, and irrigation water use efficiency is converted into a saturation-type reward term. The three parts are combined to obtain the ecological benefit index.

[0041] Finally, the yield and ecological benefit indicators for each candidate drip irrigation mode were obtained.

[0042] The detailed report on the evaluation and optimization of drip irrigation models is as follows:

[0043] Specific control parameters are read from the optimal drip irrigation mode, including the amount of water applied per irrigation, the irrigation interval, the amount of nitrogen fertilizer applied, the amount of phosphorus fertilizer applied, and the amount of potassium fertilizer applied.

[0044] The report outputs the expected yield and ecological benefits under the final optimized drip irrigation model, as well as key auxiliary indicators, including soil salinity accumulation, nitrate nitrogen leaching, irrigation water use efficiency, ecological risk standard deviation, and corresponding decision preference parameter values, forming a complete comprehensive evaluation report of the drip irrigation model.

[0045] On the other hand, the present invention also provides a drip irrigation pattern evaluation and optimization system that integrates an ecological model, including a module for executing processing instructions for each step in a drip irrigation pattern evaluation and optimization method that integrates an ecological model;

[0046] The data acquisition and preprocessing module is used to collect multi-source data from irrigated plots, complete data labeling, standardization processing and outlier correction, and output a standardized multi-source data matrix.

[0047] The dynamic collaborative feature construction module is used to perform dynamic temporal alignment on standardized data, construct an adaptive kernel function with soil salinity constraints, and generate a dynamic collaborative feature matrix through kernel principal component analysis;

[0048] The particle swarm optimization module, which incorporates ecological constraints, is used to perform iterative optimization by introducing ecological model constraints, adaptive inertia weights, and ecological correction terms, and outputs a Pareto optimal solution set.

[0049] The multi-objective comprehensive evaluation module is used to construct a comprehensive evaluation system that includes yield, ecological benefits, and uncertainty penalties to screen the optimal drip irrigation mode;

[0050] The closed-loop control and output module is used for field verification and correction, intelligent platform integration and rolling optimization, and outputs customized solutions and application reports.

[0051] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. The above technical solutions have the following advantages or beneficial effects:

[0052] This invention discloses a method and system for evaluating and optimizing drip irrigation patterns based on an integrated ecological model. It constructs a dynamic collaborative feature building method that integrates dynamic time warping and adaptive multi-kernel mapping. By introducing a soil salinity attenuation term to correct similarity calculations, it addresses the problems of inconsistent sampling frequencies, response lags, and difficulty in preserving nonlinear coupling relationships among irrigation, crop, and ecological multi-source data, enabling subsequent optimization to make decisions based on the collaborative state under ecological constraints. Furthermore, it innovatively introduces an ecological correction term into the particle swarm optimization framework, calculating the impact of each decision variable on soil salinity accumulation through finite difference sensitivity analysis, and dynamically adjusting the correction intensity in conjunction with yield status, directly driving particle reduction along the ecological risk at the velocity update level. Directional shifts have changed the traditional optimization search model that relies solely on numerical objective functions. A multi-objective comprehensive evaluation system integrating diminishing marginal benefits of yield, nonlinear mapping of ecological benefits, and uncertainty penalties has been established. The hyperbolic tangent function is used to quantify the diminishing marginal benefits of high-yield areas, and the standard deviation of ecological risk is introduced as a penalty term through Monte Carlo simulation, so that the final recommended scheme takes into account both target performance and robustness. A closed-loop application architecture from offline optimization to online adaptive control is adopted. Model parameters are corrected through small-scale validation experiments, and the rolling optimization mechanism dynamically responds to environmental changes. The construction of dynamic collaborative features and the ecological constraint optimizer are embedded in the intelligent control platform to achieve continuous linkage and iterative improvement between optimization results and the actual field system. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0054] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0055] Figure 2 Plot of variance contribution rate and cumulative contribution rate for kernel principal component analysis of dynamic collaborative characteristics of drip irrigation mode.

[0056] Figure 3 Pareto optimal frontier diagram for multi-objective optimization of drip irrigation model, focusing on yield-ecological benefits.

[0057] Figure 4 A schematic diagram illustrating the five key parameters for ultimately optimizing the drip irrigation model. Detailed Implementation

[0058] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0059] Example 1

[0060] like Figure 1As shown, a method for evaluating and optimizing drip irrigation patterns that integrates ecological models includes:

[0061] Collect multi-source data from irrigated plots and perform standardized processing;

[0062] The standardized data is dynamically aligned temporally and an adaptive kernel function is constructed to build a dynamic collaborative feature matrix.

[0063] Introducing ecological model constraints, adaptive inertia weights, and ecological correction terms into the particle swarm optimization framework, a particle swarm optimizer integrating ecological model constraints is constructed. Combining the dynamic collaborative feature matrix and crop growth model simulation results, the search direction of the particle swarm optimizer is normally corrected.

[0064] After particle swarm optimization, Pareto optimal solution is obtained, a comprehensive evaluation system is constructed, and the optimal drip irrigation mode is selected from the Pareto optimal solution;

[0065] The optimal drip irrigation mode is evaluated and optimized. After field verification and correction, the optimal drip irrigation mode is integrated into the intelligent control platform. A rolling optimization mechanism is introduced for dynamic adaptive regulation and supports customized solution output. Finally, a drip irrigation mode evaluation and optimization application report is generated.

[0066] In a specific implementation method, data collection is carried out as follows:

[0067] By deploying various monitoring devices and experimental testing methods in typical application areas of drip irrigation systems, multi-source raw data covering multiple irrigation cycles or complete growth stages are collected. The collected data samples are systematically labeled to provide accurate and structured input for subsequent dynamic collaborative feature construction, optimizer training and comprehensive evaluation.

[0068] Data acquisition begins with the irrigation control equipment in the drip irrigation system. Irrigation parameters for each irrigation event within the target monitoring period are obtained for each plot through automatic recording or timed export. These parameters include irrigation volume, irrigation interval, nitrogen fertilizer application rate, phosphorus fertilizer application rate, and potassium fertilizer application rate. This data is stored as event logs, with sampling frequency typically based on daily or the time of irrigation event occurrence. Simultaneously, crop growth parameters are periodically collected using field monitoring equipment. At typical observation points, crop height, stem diameter, leaf area index, aboveground biomass, leaf chlorophyll content, and canopy temperature are measured according to preset growth stages or fixed time intervals. Plant height, stem diameter, and canopy temperature can be measured on-site using portable instruments; leaf area index and aboveground biomass can be obtained through sampling and laboratory analysis; and leaf chlorophyll content is measured using a chlorophyll meter. The collection frequency of these crop parameters is typically lower than the irrigation event recording frequency, forming a sparse time series. In addition, the collection of ecological indicators is mainly completed through field sampling and experimental testing. At key growth stages or fixed time points, surface and root soil samples are collected to measure soil salinity, soil nitrate nitrogen content, and soil moisture content. At the same time, groundwater depth is recorded in conjunction with groundwater monitoring wells. The sampling frequency of ecological indicators is usually based on important time nodes, forming multi-source data that are unevenly distributed on the time axis with irrigation parameters and crop growth parameters.

[0069] During data collection, all observation records for each plot within a complete irrigation cycle or a complete growth stage are defined as a sample, with the total number of samples denoted as N. To meet the training data requirements of subsequent models, complete data annotations need to be constructed for each sample. The annotation process first determines the basic information of the plot corresponding to each sample, including fixed attributes such as plot location, soil type, crop variety, and planting density. Second, a unified time index is established for the sample, dividing the entire monitoring period into multiple consecutive growth stages or standardized time points. The scattered irrigation parameters, crop growth parameters, and ecological indicators are associated with the time index according to their actual observation time points. For stages with missing observations in the time index, they are filled in by linear interpolation of adjacent observations or estimation based on growth patterns, ensuring that each sample has complete parameter records in each stage, forming a labeled dataset.

[0070] The annotation categories are mainly divided into three types: The first type is irrigation pattern annotation, which is the sequence of irrigation operation parameters actually performed by the sample during the monitoring period, including the amount of water per irrigation, irrigation interval, and application amount of nitrogen, phosphorus and potassium fertilizers, which serves as the reference basis for the subsequent optimizer; the second type is crop response annotation, which is the sequence of crop growth parameters actually observed by the sample during the corresponding growth stage, reflecting the crop growth status under different irrigation patterns; the third type is ecological feedback annotation, which is the sequence of ecological indicators actually monitored by the sample during the corresponding period, focusing on recording the dynamic changes of ecological risk-related indicators such as soil salinity and nitrate nitrogen content.

[0071] In a specific implementation, the process of constructing the dynamic collaborative feature matrix is ​​as follows:

[0072] Because irrigation parameters, crop growth parameters, and ecological indicators in drip irrigation systems originate from irrigation control equipment, field monitoring equipment, and experimental testing, these different data sources often differ in sampling period, unit of measurement, noise level, and record length. This can easily lead to problems where multiple data points corresponding to the same growth stage cannot be directly aligned. Conventional processing methods often employ uniform time interpolation or simple mean filling, which, while producing a formally complete data table, can easily disrupt the inherent nonlinear coupling between irrigation behavior, crop response, and ecological feedback. This makes it difficult for subsequent optimization models to accurately identify the true correlation between high yield and low ecological risk. This invention transforms multi-source heterogeneous data into a unified dynamic collaborative feature matrix through standardization, dynamic temporal alignment, and adaptive multi-kernel mapping. The dynamic collaborative feature matrix simultaneously retains irrigation operation information, crop response information, and ecological feedback information, and introduces soil salinity process constraints. This allows the subsequent optimizer to no longer rely solely on static single indicators but to perform search and decision-making based on the multi-source collaborative state. The specific steps are as follows:

[0073] (1) The present invention first standardizes each type of raw data to make each type of parameter comparable in subsequent time alignment and similarity calculation. The specific steps are as follows:

[0074] 1) Collect the original irrigation parameter matrix, the original crop growth parameter matrix, and the original ecological indicator matrix. Specifically,

[0075] The original irrigation parameter matrix is ​​denoted as The size is , used to represent One sample in Observations on irrigation parameters;

[0076] The original crop growth parameter matrix is ​​denoted as The size is , used to represent One sample in Observed values ​​of crop growth parameters;

[0077] The original ecological index matrix is ​​denoted as The size is , used to represent One sample in Observed values ​​on ecological indicators;

[0078] in, This represents the total number of samples. Each sample can correspond to the observation records of a plot of land during a complete irrigation cycle or a complete growth stage. An example value is 120. This indicates the number of irrigation parameter types, with a value of 5. This indicates the number of crop growth parameters, with a value of 6. This indicates the number of ecological indicator types, with a value of 4.

[0079] In one embodiment, irrigation parameters include: irrigation volume (mm), irrigation interval (days), nitrogen fertilizer application rate (kg / ha), phosphorus fertilizer application rate (kg / ha), and potassium fertilizer application rate (kg / ha); crop growth parameters include: plant height (cm), stem diameter (mm), leaf area index (LAI), aboveground biomass (kg / ha), leaf chlorophyll content (SPAD value), and canopy temperature (°C); ecological indicators include: soil salinity (g / kg), soil nitrate nitrogen content (mg / kg), soil moisture content (m³ / m³), and groundwater depth (m).

[0080] 2) The original irrigation parameter matrix, the original crop growth parameter matrix, and the original ecological index matrix are standardized by column. In practice, the standardization by column first calculates the mean and standard deviation of the same type of parameter in the entire sample range, and then subtracts the corresponding mean from each observation value in the column and divides it by the corresponding standard deviation, so as to obtain the standardized irrigation parameter matrix, the standardized crop growth parameter matrix, and the standardized ecological index matrix with unified dimensions.

[0081] In engineering implementation, to avoid numerical amplification due to excessively small standard deviations, a very small stabilizing term is introduced into the denominator, for example, it can be taken as... .

[0082] Furthermore, an outlier consistency check is performed on the standardized results. When the standardized result of a certain parameter on a certain sample significantly exceeds the set range (e.g., the absolute value is greater than the preset value), the observation point is marked as an outlier observation point. Further, for outlier observation points, data interpolation of adjacent periods of the same sample, replacement with the median of similar plots, or replacement with the corrected value after manual verification based on the sampling log can be used to reduce the interference of equipment failure and sudden sampling errors on subsequent dynamic time series alignment.

[0083] 3) Output standardized irrigation parameter matrix Standardized crop growth parameter matrix and standardized ecological indicator matrix ,in, Characterizes the irrigation operation features after dimensional unification. Characterize the crop response features after dimensionless measurement. Characterizes the ecological feedback features after dimensional unification.

[0084] (2) This invention first uses the dynamic time warping method to complete the multi-source time series alignment, and then constructs an adaptive kernel function that introduces ecological process decay on the alignment result to characterize the nonlinear cooperative relationship between different samples. The specific steps are as follows:

[0085] 1) Establish a unified timeline for the growth stages to provide a common reference coordinate for data from different sources. Specifically, based on the growth patterns of the target crop, the entire growth period is divided into a unified sequence of stages. For example, it can be divided into the emergence stage, seedling stage, vigorous growth stage, flowering and fruiting stage, and maturity stage, or it can be divided at equal intervals. There are 10 standard time points, among which... This indicates the number of stage points on a unified timeline. The value can be set according to the monitoring accuracy, for example, it can be 10.

[0086] Furthermore, based on a unified timeline of reproductive stages, the standardized irrigation parameter matrix, standardized crop growth parameter matrix, and standardized ecological indicator matrix are reorganized into a sample-level time-series vector. Specifically,

[0087] For the first A sample was recombined to obtain a time-series vector of irrigation parameters. Crop growth parameter time series vector and ecological indicator time series vector ,in, Characterizing the first Irrigation parameter sequences for a sample over a unified reproductive stage timeline. Characterizing the first The crop growth parameter sequence of each sample on a unified reproductive stage timeline. Characterizing the first The ecological indicator sequence of a sample over a unified reproductive stage timeline.

[0088] 2) Perform dynamic time warping on multiple time series vectors within the same sample. Specifically, establish cost matrices for irrigation parameter time series vectors and crop growth parameter time series vectors, irrigation parameter time series vectors and ecological indicator time series vectors, and crop growth parameter time series vectors and ecological indicator time series vectors, respectively. Each cell in the cost matrix is ​​used to represent the degree of difference between two stage points. This degree of difference can be calculated using Euclidean distance, absolute difference, or multidimensional vector distance. Then, use dynamic programming to search for the alignment path with the minimum cumulative cost to obtain the optimal stage correspondence between different data sources.

[0089] For example, suppose there are two time series vectors, the irrigation parameter time series vector. (3 stage points), time-series vector of crop growth parameters (4 stage points), the cost matrix is ​​a A matrix, where each element express and The degree of difference between them. During construction, first create an empty matrix with [number of rows]. The number of columns is Then iterate through each row index. (from 1 to 3) and column indexes (From 1 to 4); then calculate and The distance, if and If it is a single value, calculate the absolute difference. If they are multidimensional vectors (e.g., a stage point contains multiple parameters), then calculate the Euclidean distance; finally, fill the calculation results into the corresponding positions in the matrix. .

[0090] 3) Based on the alignment path obtained from dynamic time warping, the original unequal-length sequences are mapped onto a unified reproductive stage timeline. Specifically,

[0091] For a one-to-one corresponding stage point, the parameter value of that stage point is directly retained; for many-to-one or one-to-many corresponding stage points, the parameter value of the unified stage point is generated by using local mean, weighted mean, or linear interpolation.

[0092] Based on this, the time series vectors of irrigation parameters, crop growth parameters, and ecological indicators for each sample are kept consistent in length, while preserving the true response relationship caused by the stage misalignment as much as possible.

[0093] 4) Extract soil salinity time-series data from the ecological indicator time-series vector and construct a soil salinity constraint sequence. Specifically, for the first... The soil salinity time-series data for each sample are denoted as follows: , used to indicate the first The soil salinity changes of a sample at multiple monitoring times, with a length of [duration missing]. ,in, This indicates the number of time points for soil salinity monitoring; an example value is 20.

[0094] In practical implementation, because the ecological indicator time series vector It is itself a sequence on a unified reproductive stage timeline, containing various ecological indicators such as soil salinity and nitrate nitrogen. The extraction process is equivalent to indexing based on the predefined feature of "soil salinity" in a vector. For example, assuming... It is The matrix ( Each stage If the data is classified as ecological indicators and the column order is fixed as [soil salinity, soil nitrate nitrogen content, soil moisture content, groundwater depth], then the time series data of soil salinity will be extracted. It means taking the first column of this matrix to form a column of length . The vector.

[0095] 5) Define the timeline after dynamic time warping, remapped to a unified reproductive stage timeline. , and These are the aligned sample-level time-series vectors. All three vectors have the same length. Furthermore, each stage corresponds to a specific point in time, reflecting the optimal alignment of irrigation behavior, crop response, and ecological feedback over time.

[0096] Furthermore, an adaptive kernel function is constructed using the aligned sample-level time-series vector as input. Specifically, multiple basic kernel functions are first set based on the fundamental similarity between samples, and then soil salinity differences are introduced to form an ecological process attenuation term.

[0097] Among them, soil salinity difference refers to the difference between two samples (with the first sample as the first sample). The first sample and the first (Taking a sample as an example) Soil salinity time-series data on their respective unified reproductive stage timelines. and The degree of overall difference between them quantifies the similarity of salinity dynamics between two plots or two irrigation cycles throughout the growing season.

[0098] In practical implementation, the number of basis kernel functions can be three, with two using Gaussian kernels of different bandwidths and one using a Laplace kernel. These different basis kernel functions are used to characterize local similarity, mesoscale similarity, and abrupt change sensitivity, respectively. The fusion weight of each basis kernel function is denoted as... If the sum of all weights is 1, the weights can be initialized to equal values ​​first, and then updated using cross-validation, the minimum prediction error criterion, or the validation set performance maximization criterion.

[0099] In practical implementation, soil salinity differences are introduced to form an ecological process attenuation term. Specifically, a function is used to measure the temporal difference in soil salinity between two samples. Mapped to a decay coefficient between 0 and 1, this function can be a monotonically decreasing function, such as an exponential decay function. ,but The larger the value (the greater the difference in the salinization process), the smaller the attenuation coefficient (close to 0). The smaller the value (the more similar the salinization processes), the larger the attenuation coefficient (close to 1). Therefore, for any two samples, first calculate the basic similarity based on the aligned multi-source features, then construct an ecological process attenuation term based on the temporal differences in soil salinity between the two samples, and apply ecological process attenuation to avoid misidentifying two samples with "similar parameters but significantly different salinization processes" as highly similar samples. For ease of implementation, the adaptive kernel function value can be obtained by multiplying the multi-kernel fusion result by the ecological process attenuation term, expressed as: ;in, The item represents the basic similarity. The term is the ecological process decay term, and the product of the two terms yields the final adaptive kernel function value; This represents the adaptive kernel function value, used to quantify the overall similarity between two samples under ecological constraints; Indicates the first The similarity of the outputs of each basis kernel function; This represents the ecological constraint strength coefficient, used to control the strength of the influence of soil salinity differences on the final similarity. An example value is 0.5. The difference between two samples of soil salinity time series data can be calculated using Manhattan distance.

[0100] It should be noted that the ecological process decay term is used to correct the basic similarity calculated by the basic kernel function. Its purpose is to reduce the final similarity between two samples that, although irrigation and crop responses are similar, have very different soil salinity accumulation processes. This forces the optimizer to recognize that even if irrigation and crop indicators appear similar, they should be considered different states if the accompanying salinity risk patterns are different, thereby avoiding recommending high-risk irrigation patterns.

[0101] It should also be noted that crop response and ecological feedback in drip irrigation systems usually lag behind irrigation events. If this lag is ignored, the model will misjudge the true coupling relationship as noise. At the same time, if the differences in soil salinity processes are ignored, the model is prone to mixing ecological risk patterns with low-risk patterns. This invention does not simply splice different data sources directly at the same time, but first performs dynamic temporal alignment based on the stage response relationship, and then introduces the differences in ecological processes as a constraint term in the kernel function into the similarity calculation. Based on this, the dynamic coupling relationship between irrigation behavior, crop response and ecological processes can be preserved more accurately.

[0102] (3) This invention constructs an adaptive kernel matrix based on an adaptive kernel function, and then uses kernel principal component analysis to extract low-dimensional dynamic collaborative features, so as to reduce the computational complexity of subsequent optimization while retaining nonlinear coupling information. The specific steps are as follows:

[0103] 1) Regarding the first For each sample, the aligned irrigation parameter time-series vector, crop growth parameter time-series vector, and ecological indicator time-series vector are concatenated in a fixed order to obtain the full parameter concatenation vector. Fully parameterized concatenation of vectors Characterizing the first The complete multi-source feature representation of each sample includes all irrigation operation information, crop response information, and ecological feedback information of the sample on a unified growth stage time axis.

[0104] It should be noted that the splicing order is kept consistent across all samples to ensure the comparability of kernel value (adaptive kernel function value) calculations.

[0105] 2) For any two samples (let ) The first sample and the first Taking a sample as an example, the kernel value between them is calculated using an adaptive kernel function, and all kernel values ​​are written into the adaptive kernel matrix, denoted as . The size is Adaptive kernel matrix The element in the i-th row and j-th column Characterizing the first The sample and the first The nonlinear collaborative similarity between samples indicates that the larger the kernel value, the closer the two samples are in terms of irrigation operations, crop response, and ecological feedback.

[0106] 3) The adaptive kernel matrix is ​​centered to eliminate the influence of the overall mean on the principal component extraction results, so that the kernel principal component analysis focuses more on the relative differences between samples rather than the absolute shift.

[0107] In practical implementation, a two-sided mean-reduction method can be used, that is, subtracting the row mean and column mean respectively and adding back the overall mean to obtain the centered kernel matrix.

[0108] Furthermore, eigenvalue decomposition is performed on the centered kernel matrix to obtain the kernel principal component directions sorted by contribution. Eigenvalue decomposition is performed on the centered kernel matrix. Perform eigenvalue decomposition, specifically for Symmetric matrix Decompose it into In the form of, yes An orthogonal matrix, where each column is an eigenvector. It is a diagonal matrix, and the elements on the diagonal are the corresponding eigenvalues. In kernel principal component analysis, these eigenvectors define new eigenspace directions, and the magnitude of the eigenvalues ​​represents the variance of the data distribution in that direction, i.e., the importance of the principal component.

[0109] Then, according to the preset rules, the first few principal components are retained to obtain the dynamic collaborative feature dimensions. Dynamic collaborative feature dimensions The number of dynamic collaborative features that are ultimately retained can be set using a fixed dimension, such as 50, or using a cumulative contribution rate, such as stopping retention when the cumulative contribution rate reaches 90% or 95%.

[0110] 4) Project each sample onto the direction of the retained principal components to obtain the dynamic collaborative feature matrix. Dynamic collaborative feature matrix The size is Each row corresponds to a low-dimensional collaborative feature representation of a sample, and each column corresponds to a dynamic collaborative feature component extracted by kernel principal component analysis.

[0111] Dynamic collaborative feature matrix The particle swarm optimizer, which is used for subsequent integration of ecological model constraints, can directly read the dynamic collaborative feature vector of the corresponding sample as the model state input for single-plot optimization scenarios. For regional optimization scenarios, the dynamic collaborative feature matrix can be clustered according to plot type, soil type or management zone, and then the cluster center or weighted average vector can be used as the representative state input.

[0112] like Figure 2 The analysis of the eigenvalue contribution rate in principal component analysis is described above. Figure 2The bar chart represents the variance contribution rate of each kernel principal component, and the line chart represents the cumulative contribution rate, with the 90% cumulative contribution threshold marked. The horizontal axis represents the "kernel principal component number," and the vertical axis represents the "variance contribution rate (%)." Kernel principal component analysis can compress high-dimensional nonlinear collaborative features into a low-dimensional space, and the first few principal components can retain most of the information, demonstrating the effectiveness of dimensionality reduction in dynamic collaborative feature construction. It preserves nonlinear coupling relationships while reducing the computational complexity of subsequent optimizers.

[0113] In a specific implementation, the construction process of the particle swarm optimizer is as follows:

[0114] (1) Particle encoding and initialization

[0115] 1) Define the drip irrigation mode parameter vector as the particle encoding method. Specifically,

[0116] The particle position vector is denoted as , characterizing the The particle in the first A corresponding set of candidate drip irrigation mode parameters; Represents an iterative algebra, with values ​​ranging from 0 to... ; This represents the particle index, with values ​​ranging from 1 to... ;

[0117] The dimension of the drip irrigation mode parameter vector is denoted as . The value can be 5, corresponding to the amount of water applied each time, the watering interval, the amount of nitrogen fertilizer applied, the amount of phosphorus fertilizer applied, and the amount of potassium fertilizer applied, respectively. Optionally, in the case of a fixed formula ratio, the amount of phosphorus fertilizer applied and the amount of potassium fertilizer applied can also be converted according to a preset ratio, thereby simplifying the drip irrigation mode parameter vector to 3 dimensions. The default setting of the drip irrigation mode parameter vector dimension in this invention is 5.

[0118] Furthermore, the range of values ​​for each dimension parameter is set according to agronomic and equipment constraints to limit the particle search space and prevent particles from entering obviously unreasonable agricultural operation areas. For example, the amount of water applied each time can be set between 10 mm and 60 mm, the irrigation interval can be set between 2 days and 12 days, and the application rates of nitrogen fertilizer, phosphorus fertilizer, and potassium fertilizer can be set with upper and lower limits according to the recommended fertilization range for the target crop.

[0119] 2) Initialize the particle swarm, with the particle swarm size denoted as . The example value is 50. For each particle, an initial position vector is randomly generated within the allowable range of each dimension parameter, and an initial velocity vector is also generated.

[0120] The velocity vector is denoted as This is used to control the direction and magnitude of particle movement in the parameter space. To avoid an overly drastic initial search, the initial velocity can be set to 10% to 20% of the range of values ​​for each parameter.

[0121] 3) Input the corresponding state vector in the dynamic collaborative feature matrix and the drip irrigation mode parameter vector of each particle into the model evaluation module. The model evaluation module includes at least a soil salinity balance model, a crop growth model and a nitrogen balance model. Based on this, the evaluation results of the same set of drip irrigation parameters under different ecological backgrounds and crop conditions can reflect environmental differences, rather than using a uniform score that is detached from the scenario.

[0122] In practical implementation, the model evaluation module takes as input the drip irrigation mode parameter vector for each particle. Dynamic collaborative feature vectors corresponding to the scenario (from The system obtains initial soil and crop state parameters and outputs soil salinity balance model (outputs root zone soil salinity accumulation, salt leaching, etc.), crop growth model (outputs crop yield, biomass, growth status at each growth stage, etc.), and nitrogen balance model (outputs nitrate nitrogen leaching, soil nitrogen residue, etc.).

[0123] In one implementation, the soil salinity balance model, crop growth model, and nitrogen balance model employ mature models from the agricultural environment field. Examples include soil water and salt transport models based on the Richards equation, evapotranspiration calculations using the FAO-56 dual crop coefficient method, crop growth models based on platforms such as DSSAT or APSIM, and nitrogen balance models based on nitrogen cycle processes.

[0124] The soil salinity balance model is a process model built on the principles of water balance and salt transport. It simulates the salt brought in by irrigation water and rainfall, the salt absorbed by crops, and the salt leached out by deep seepage, thereby dynamically calculating the changes in soil salinity in the root zone and quantifying the risk of soil salinity accumulation caused by different drip irrigation modes.

[0125] Crop growth models are mechanistic models (such as DSSAT, APSIM, etc.) or simplified empirical models based on physiological and ecological processes. They simulate the entire process of crops from emergence to maturity based on meteorological, soil, irrigation and fertilization conditions, including photosynthesis, respiration, and material allocation, and are used to predict crop yield and biomass under different drip irrigation modes.

[0126] The nitrogen balance model is a process model that simulates the transformation (mineralization, nitrification, denitrification), absorption and transport of nitrogen in soil. It calculates the applied nitrogen fertilizer, the initial nitrogen in the soil, the nitrogen absorbed by the crop, and the nitrogen lost through leaching or denitrification, and is used to quantify the risk of nitrate nitrogen leaching caused by different drip irrigation modes.

[0127] 4) Based on the initial evaluation results, set the individual historical best position of each particle and establish an external non-dominated solution archive. The external non-dominated solution archive is an independent and continuously updated storage space used to save all non-dominated good solutions (i.e., Pareto optimal solutions) found so far in the optimization process. It is independent of the particle swarm itself. Particles in the particle swarm only record their own historical best positions, while this archive maintains the global optimal solution frontier in the entire search space. When subsequent particles are updated, the global guiding position will be selected from this archive.

[0128] In practice, the initial evaluation results are calculated by the model evaluation module. That is, after the particle position is initialized, the soil salinity balance model, crop growth model and nitrogen balance model are immediately called to simulate and evaluate the initial drip irrigation mode parameters of each particle, and obtain the initial yield, ecological risk and other indicators. These indicators constitute the initial evaluation results.

[0129] (2) Particle velocity update and ecological correction

[0130] The invention adds adaptive inertia weights and ecological correction terms to the standard velocity update framework, enabling particles to automatically reduce their tendency to blindly converge when approaching high ecological risk regions and to deflect along the direction of decreasing ecological risk. The specific steps are as follows:

[0131] 1) Based on the external non-dominated solution archive, select a global guiding position for each particle, denoted as [the global guiding position is not specified in the original text]. , which represents the guided solution selected from the current external non-dominated solution archive.

[0132] In practical implementation, solutions with larger crowding distances can be prioritized to maintain the diversity of Pareto front distribution; alternatively, a roulette wheel method can be used to randomly select from non-dominated solutions to avoid premature aggregation of particle swarms in local regions.

[0133] It should be noted that crowding distance is a metric in the Pareto optimal solution set, used to measure the density of other solutions around a solution. For a solution, the crowding distance is usually defined as the sum of its distances to its two neighboring solutions in each objective dimension in the objective space. The larger the crowding distance, the sparser and more diverse the region around the solution is. When selecting a global guiding position from the external non-dominated solution archive, solutions with large crowding distances are preferred. This can guide the particle swarm to explore regions that have not yet been fully searched, thereby maintaining the diversity of the Pareto front distribution and avoiding premature convergence to a local region.

[0134] 2) Calculate the adaptive inertia weights, denoted as . It is used to control the degree to which particles inherit the velocity of the previous generation.

[0135] In the specific implementation, a basic inertia weight is first set, which gradually decreases with the number of iterations, so that the optimization is biased towards global search in the early stage and towards local refinement in the later stage; then, the inertia weight is dynamically adjusted according to the soil salinity increment corresponding to the current global guidance position; the soil salinity increment is denoted as... , representing the increase in soil salinity relative to the initial state, simulated at the current global guidance position in this round of assessment; the soil salinity safety threshold is denoted as The example value is 2.0 grams per kilogram, when the soil salinity increases. Approaching or exceeding At times, appropriately increase the inertia weight to enhance the particle's ability to escape high-risk regions; when Significantly lower At this time, the inertial weight is allowed to decrease according to the normal law, so that the particles converge to the good region more quickly.

[0136] To facilitate implementation, adaptive inertia weights can be constructed using a "linear decreasing term plus ecological feedback correction term" approach, and the results can be limited to a preset range.

[0137] Define the linearly decreasing term as It is from linearly decreasing to The basic values ​​ensure the fundamental strategies for early exploration and later development;

[0138] Define the ecological feedback correction term as ,when When, the correction term equals ,when When the correction item is proportionally less than ;

[0139] The linearly decreasing term is added to the ecological feedback correction term, and then the final adaptive inertia weight is constrained by a maximum / minimum function. Within the interval, when the salt increment at the global guiding position is too high, even in the later stages of iteration, the inertia weight will be appropriately increased to encourage particles to maintain stronger exploration capabilities in order to escape high-risk areas. This is the adaptive inertia weight. The calculation method is expressed as follows:

[0140] ;

[0141] The upper limit of the basic inertia weight is denoted as The example value is 0.9, and the lower limit of the basic inertia weight is denoted as . The example value is 0.4, and the maximum number of iterations is denoted as... The example value is 200, and the ecological feedback coefficient is denoted as... The example value is 0.2.

[0142] 3) The base velocity term is calculated based on the standard particle swarm optimization (SSO) mechanism. This base velocity term includes the inertia inherited from the previous generation's velocity, the individual learning term towards the individual's historical best position, and the swarm learning term towards the global guiding position. Specifically,

[0143] Define inertial inheritance as This represents the tendency of particles to maintain the motion trend of the previous generation;

[0144] Define individual learning items as This represents the particle's trajectory towards its historical best position. Learning inclination;

[0145] Define the group learning term as This represents the position of the particle guiding the global direction. Learning inclination;

[0146] The update method for the base velocity term is then expressed as follows:

[0147] ;

[0148] in, Indicates the first The particle in the first The fundamental velocity vector of the generation, Indicates the first The particle in the first The velocity vector of the generation, Indicates the first The particle in the first The position vector of the generation, i.e., a set of candidate drip irrigation mode parameters. This represents the individual learning factor, used to control the weights by which a particle moves closer to its historical best position; an example value of 1.5 can be used. This represents the group learning factor, used to control the weights by which particles move closer to the globally guided position; an example value of 1.5 can be used. and Let represent two independent random numbers that follow a uniform distribution in the interval [0,1]. Indicates the first The individual best position in the history of each particle. This indicates that the current iteration is the [number]th iteration. The global guiding position selected by each particle.

[0149] 4) Calculate the ecological correction term, providing a "push" for the particle along the direction of decreasing ecological risk. The ecological correction term is denoted as... , used to indicate the first To ensure feasibility, the ecological constraint correction speed of each particle in the current iteration is constructed using a method of "forward simulation + finite difference sensitivity analysis" instead of directly adopting the abstract gradient expression that is difficult to implement.

[0150] 4.1) First, input the current particle's drip irrigation mode parameter vector and the corresponding scenario's dynamic collaborative feature vector into the soil salinity balance model and crop growth model, and simulate each stage of the entire crop growth period. The number of stages in the entire crop growth period is denoted as... The example value is 5, and the stage index is denoted as . The value range is from 1 to .

[0151] 4.2) At each growth stage, apply a small perturbation to each decision variable in the particle position vector, for example, take 1% to 5% of the variable's value range, and then rerun the ecological simulation for that stage or the entire growth period. Compare the changes in soil salinity accumulation in the root zone before and after the perturbation to obtain the sensitivity of the decision variable to ecological costs.

[0152] Ecological cost is denoted as , used to indicate the first The particle in the first The amount of soil salt accumulation in the root zone caused by each growth stage, among which... Indicates the first The particle in the first The corresponding set of candidate drip irrigation mode parameter vectors. This represents the stage index during the entire growth period of a crop, with a value range of [value range missing]. arrive ;

[0153] In the actual implementation, a forward simulation is first performed, specifically for the current particle. Input the corresponding dynamic collaborative feature vector, run the soil salinity balance model and crop growth model to simulate the entire growth period, and obtain the results for each growth stage. ecological costs Then, a finite difference perturbation is performed, specifically for each decision variable in the particle position vector. Apply a small perturbation (For example, 1% of its value range), keeping other decision variables constant, resimulate this stage (or the entire cycle) to obtain the ecological cost after the disturbance. Finally, sensitivity is calculated, specifically the sensitivity of the decision variable to the ecological cost of that stage. ,but This refers to the sensitivity of the decision variable to ecological costs.

[0154] Furthermore, by combining the sensitivity levels corresponding to all decision variables, the ecological sensitivity vector for this stage is obtained. The larger the value of the ecological sensitivity vector, the more significant the impact of the corresponding parameter change on salinity accumulation.

[0155] 4.3) Perform finite difference perturbation. Specifically, adjust the ecological sensitivity vector according to the crop stage and yield status. The simulated crop yield or yield contribution at each stage is denoted as... The benchmark output is denoted as , used to represent the target yield level under conditions of no significant salt stress;

[0156] When the current stage yield is significantly lower than the baseline yield, the intensity of ecological correction is reduced to prevent the optimizer from excessively suppressing irrigation and fertilization, which would cause the yield to continue to decline. When the current stage yield is close to the baseline yield or higher than the acceptable range, the intensity of ecological correction is increased so that the optimizer can prioritize reducing ecological risks while ensuring yield.

[0157] In practical implementation, the ecology-yield coupling regulation coefficient is denoted as... The example value is 0.05; the ecological correction step size is denoted as... The example value is 0.1, which is adjusted through the ecology-yield coupling coefficient. and benchmark production The sensitivity vector is adjusted to generate a modified intensity factor. , is represented as:

[0158] ;

[0159] When simulated output Below benchmark production hour, If it is a negative value, then The term is 0, therefore The ecological correction term is weakened; when the simulated yield is higher than the baseline yield, the value is positive, and the higher the yield, the stronger the correction.

[0160] Based on this, the direction of the ecological correction term is determined by the sign of the sensitivity, while the step size of the correction is determined by... The ecological correction term is determined by both the magnitude of the sensitivity and the specificity of the sensitivity. It can be represented as The vector form represents the direction that propels particles toward reducing salt accumulation, where... Represents a symbolic function.

[0161] 4.4) The adjusted ecological sensitivity vectors from each stage are summed to form the ecological correction term for the current particle, and this ecological correction term is applied to the particle velocity along the direction of decreasing ecological cost. In other words,

[0162] When increasing a decision variable would significantly exacerbate soil salinity accumulation, the ecological correction term will push particles to reduce that variable; when adjusting a decision variable helps mitigate salinity risks without significantly sacrificing yield, the ecological correction term will push particles to move in that direction.

[0163] 5) Add the basic velocity term to the ecological correction term to obtain the new particle velocity, and set an upper limit for the velocity of each dimension to prevent the particle from jumping out of the feasible region directly due to excessive leap in a certain round of update. The velocity upper limit can be set to 20% to 30% of the width of the corresponding parameter value range.

[0164] It should be noted that agricultural ecological processes often exhibit significant path dependence. For example, once salt accumulation occurs continuously, it is difficult to fully recover through local corrections later. Therefore, it is necessary to pull particles back from high-risk directions as early as possible during the search process. The ecological correction term is not simply a static penalty coefficient deducted from the objective function, but rather the particle's direction of motion is directly changed during the velocity update phase. Based on this, the optimizer can simultaneously consider search efficiency and ecological safety, reducing the occurrence of results that are "numerically optimal but unusable in the field".

[0165] (3) Particle position update and feasible region correction

[0166] 1) Iteratively update the particle position based on the updated particle velocity, and denote the updated particle position as . , used to indicate the first The parameter vector of the candidate drip irrigation mode corresponding to each particle in the next generation.

[0167] In practice, the current particle position can be added to the updated particle velocity dimension by dimension to obtain the new particle position.

[0168] 2) Perform feasible region correction on components that exceed the upper and lower limits of the parameters. Specifically,

[0169] If a parameter exceeds the upper limit, it is truncated to the upper limit, or a mirror bounce method is used to reflect the excess part back to the allowable range. When using the mirror bounce method, the corresponding velocity component is inverted and multiplied by the attenuation coefficient to reduce oscillations at the boundary. If a parameter is below the lower limit, the lower limit is truncated or bounce correction is performed in the same way.

[0170] 3) Conduct combined checks on agronomic constraints, for example,

[0171] When the amount of water applied each time is large, the interval between waterings should not be too short at the same time; the total amount of nitrogen fertilizer, phosphorus fertilizer, and potassium fertilizer applied should not exceed the upper limit allowed for the corresponding stage; in scenarios where the fertilizer ratio is limited, the amount of nitrogen fertilizer, phosphorus fertilizer, and potassium fertilizer applied should meet the preset ratio.

[0172] If the new position does not satisfy the combined constraints, then project it onto the feasible solution closest to the current point.

[0173] 4) Use the updated particle positions to re-invoke the model evaluation module to obtain new production target values, ecological target values ​​and related constraint states. Then, update the individual's historical best position according to the Pareto dominance relationship: if the new position is better than the current individual's historical best position in at least one objective and not inferior in the other objectives, then replace it with the new position; if the two do not dominate each other, then the position with a more even distribution or a lower degree of constraint violation is preferred to be retained.

[0174] It should be noted that the model evaluation module (including soil salinity balance, crop growth, and nitrogen balance models) calculates new indicators such as yield, salt accumulation, and nitrogen leaching based on the input drip irrigation mode parameters (such as irrigation volume and fertilizer application) and the current environmental / crop status. When the particle position (i.e., drip irrigation mode parameters) is updated, these new parameters are re-inputted into these models, and the models will simulate the corresponding new output results.

[0175] 5) Update the external non-dominated solution archive based on the latest evaluation results of all particles, delete dominated solutions, and retain new non-dominated solutions; when the archive capacity reaches the upper limit, solutions in overly dense regions can be deleted based on the crowding distance to ensure the uniform distribution of the non-dominated solution set in the target space.

[0176] 6) Repeat the particle velocity update, particle position update, and archive update until the maximum number of iterations is reached. If the external non-dominated solution archive changes very little in multiple consecutive iterations, the optimization process is considered to have converged.

[0177] In a specific implementation, the comprehensive evaluation and output of the drip irrigation model based on the multi-objective evaluation system are as follows:

[0178] (1) Obtaining the Pareto optimal solution set

[0179] 1) After particle swarm optimization is completed, all non-dominated solutions are extracted from the external non-dominated solution archive to form a Pareto optimal solution set, denoted as . , which includes One candidate drip irrigation mode This represents the number of Pareto optimal solutions, and its value is determined by the optimization results.

[0180] Furthermore, the first Pareto optimal solution in the set... The parameter vector of the candidate drip irrigation mode is denoted as . , characterizing the The specific values ​​of each candidate drip irrigation mode in the parameter space, and the dimension of the drip irrigation mode parameter vector. Consistent (default setting is 5).

[0181] 2) Perform deduplication and nearest neighbor merging on the Pareto optimal solution set. Specifically, if the difference between two candidate drip irrigation modes in each dimension of parameters is less than the set tolerance, for example, less than 1% to 2% of the allowable range of the corresponding parameters, then the one with more stable model evaluation can be retained to avoid a large number of almost duplicated schemes in the subsequent comprehensive evaluation.

[0182] In one implementation, the multi-objective evaluation system is a multi-objective optimization problem that includes one economic benefit (yield) and multiple ecological benefits (salinity, nitrogen leaching, water efficiency). Specifically, it includes: maximizing yield benefits (such as simulating total yield). (kg / ha) and minimizing ecological risk, where minimizing ecological risk is itself a comprehensive objective, including: minimizing soil salinity accumulation. (g / kg), Minimum nitrate nitrogen leaching loss (kg / ha) and maximizing irrigation water use efficiency (kg / m³).

[0183] (2) Calculation of output benefits and ecological benefits

[0184] 1) For each candidate drip irrigation mode in the Pareto optimal solution set, the crop growth model, soil salinity balance model and nitrogen balance model are called for forward simulation. During the simulation, the parameter vector of the candidate drip irrigation mode and the dynamic collaborative feature vector of the corresponding scenario are used as input. The dynamic collaborative feature vector is used to provide comprehensive state information of soil, crop and ecological background under the scenario, so that the evaluation results can reflect the real field conditions.

[0185] Based on the crop growth model, the yield benefit index of this candidate drip irrigation mode is output (one of the outputs of the crop growth model is total yield). The yield benefit index of the candidate drip irrigation modes is marked as The unit is kilograms per hectare, representing the simulated total yield of the candidate model in the target scenario. The larger the value, the stronger the agricultural output capacity;

[0186] The output of the soil salinity balance model is as follows: The soil salinity accumulation of each candidate drip irrigation pattern (the soil salinity balance model can simulate and output the root zone soil salinity accumulation at the end of the entire growing season) is denoted as The unit is grams per kilogram;

[0187] Based on the nitrogen balance model, the output is... The nitrate nitrogen leaching loss of each candidate drip irrigation mode (one of the outputs of the nitrogen balance model is the total amount of nitrate nitrogen leached from the root zone through deep percolation) is denoted as: The unit is kilograms per hectare;

[0188] The irrigation water use efficiency of this candidate drip irrigation model is calculated based on total output and total irrigation water volume (using total output). Dividing by the total irrigation water volume (this is a derived indicator), denoted as The unit is kilograms per cubic meter.

[0189] 2) Soil salinity accumulation, nitrate nitrogen leaching, and irrigation water use efficiency are converted into ecological benefit indicators, which are denoted as follows: A higher value indicates better ecological performance. Specifically,

[0190] First, the accumulation of soil salinity and the leaching of nitrate nitrogen are converted into attenuation terms, where, This represents the salt decay term. This represents the nitrogen leaching attenuation term; the higher the risk, the stronger the attenuation. Irrigation water use efficiency is then converted into a saturation-type reward term. In other words, higher efficiency leads to greater rewards, but the benefits of further improvements gradually decrease once efficiency is already high. Finally, the three components are combined to obtain the ecological benefit index. ;

[0191] in, This represents the ecological threshold for soil salinity, with an example value of 3.0 grams per kilogram. This represents the soil salinity sensitivity index, with an example value of 2.0. This represents the ecological threshold for nitrate nitrogen leaching, with an example value of 30.0 kg per hectare. This represents the sensitivity index for nitrate nitrogen leaching, with an example value of 1.5. This represents the baseline irrigation water use efficiency, with an example value of 2.5 kg per cubic meter.

[0192] Based on this, the yield benefit indicators and ecological benefit indicators corresponding to each candidate drip irrigation mode are output, forming the basic data table for subsequent comprehensive scoring.

[0193] (3) Construction and decision-making of comprehensive scoring function

[0194] 1) Statistically analyze the yield benefit index and ecological benefit index of all candidate drip irrigation modes in the Pareto optimal solution set, and obtain the minimum yield benefit, maximum yield benefit, minimum ecological benefit and maximum ecological benefit respectively;

[0195] The minimum output efficiency is denoted as The maximum output benefit is denoted as The minimum ecological benefit is denoted as The maximum ecological benefit is denoted as .

[0196] 2) Perform nonlinear utility mapping on the yield benefit index of each candidate drip irrigation mode. Specifically, first normalize the yield benefit index according to the minimum and maximum values ​​of the Pareto optimal solution set, and then feed it into the hyperbolic tangent function to obtain the yield utility value.

[0197] It should be noted that the hyperbolic tangent function is used because there is a law of diminishing marginal returns in actual agricultural decision-making regarding yield increases. Once the yield has reached a high level, pursuing a small increase in yield should generally not come at the cost of significantly increasing ecological risks. The hyperbolic tangent function can map the normalized yield to a saturated range of utility values. In low-yield areas, the yield utility value is sensitive to changes in yield, while in high-yield areas, the rate of increase in yield utility slows down. This aligns with the understanding in actual decision-making that the tolerance for "increased yield without increased efficiency" in high-yield areas is decreasing.

[0198] 3) The ecological benefit indicators of each candidate drip irrigation mode are normalized to obtain the ecological utility value. After normalization, the higher the ecological utility value, the better the overall performance of the candidate mode in terms of low salt accumulation, low nitrogen leaching and high irrigation water use efficiency.

[0199] 4) Perform uncertainty analysis on each candidate drip irrigation mode to obtain the standard deviation of ecological risk, denoted as . , used to indicate the first The degree of fluctuation in ecological benefits of candidate drip irrigation models under parameter perturbation and model perturbation.

[0200] In practical implementation, Monte Carlo simulations are performed on each candidate drip irrigation mode: the irrigation volume, irrigation interval, fertilization parameters, and key ecological model parameters are set as random variables, and samples are taken near their nominal values ​​according to a preset distribution, such as a normal or uniform distribution, and the simulations are repeated. Next, get The results of the group's ecological benefits include, This indicates the number of Monte Carlo simulations, with an example value of 1000. Then, the standard deviation of the ecological risk of the candidate drip irrigation model is calculated based on all simulation results. The larger the standard deviation of the ecological risk, the more sensitive the candidate model is to uncertainty, and the higher the risk of practical application.

[0201] 5) Combine the yield utility value, ecological utility value, and ecological risk penalty item into a comprehensive score. The comprehensive score of the candidate drip irrigation modes is recorded as follows: This process can be described by the following formula:

[0202] ;

[0203] in, Indicates the first Yield utility values ​​for each candidate drip irrigation mode Indicates the first Ecological utility values ​​of candidate drip irrigation models; This represents the decision-maker's risk preference factor, with a value ranging from 0 to 1. The larger the value, the more it is biased towards production targets. The smaller the value, the more it leans towards ecological goals; This represents the ecological risk aversion coefficient, with an example value of 0.2, used to control the intensity of the ecological risk penalty term.

[0204] 6) Select the candidate drip irrigation mode with the highest comprehensive score from all candidate drip irrigation modes as the final recommended scheme;

[0205] If the overall scores of multiple candidate drip irrigation modes are very close, for example, the difference is less than a preset threshold, then the candidate drip irrigation mode with a smaller ecological risk standard deviation, more stable parameters, or easier implementation will be given priority.

[0206] It should be noted that actual irrigation decisions are driven by both yield targets and ecological risk tolerance. Different decision-makers have different preferences. At the same time, there are unavoidable uncertainties in the field environment and model parameters. Selecting a scheme based solely on the results of a single deterministic simulation can easily lead to recommendations with insufficient robustness. The final decision of this invention is not a simple fixed weighted average of yield and ecological indicators, but rather introduces both a diminishing marginal benefit mechanism and an uncertainty penalty mechanism. Based on this, the final output drip irrigation model can take into account both target performance and application stability.

[0207] (4) Final optimization of drip irrigation mode output

[0208] 1) The candidate drip irrigation mode with the highest comprehensive score is recorded as the parameter vector of the final optimized drip irrigation mode. This represents the final output drip irrigation management scheme, and its dimension is the same as the dimension of the drip irrigation mode parameter vector. Consistent (default setting is 5).

[0209] Furthermore, specific control parameters are read from the final optimized drip irrigation mode parameter vector. For example, if a 5-dimensional encoding method is used, the irrigation volume for each cycle is output. Watering interval Nitrogen fertilizer application rate Phosphate fertilizer application rate and potassium fertilizer application rate ,in, The unit is millimeters. The unit is days. , and The unit can be set to kilograms per hectare based on the actual fertilization method.

[0210] 2) Synchronously output the expected yield and expected ecological benefits corresponding to the final optimized drip irrigation mode: the final expected yield is denoted as... This represents the predicted yield of the optimized drip irrigation model under the target scenario; the final expected ecological benefit is denoted as... This indicates the overall ecological performance of the optimized drip irrigation model in terms of salt control, nitrogen leaching control, and irrigation water use efficiency.

[0211] 3) The output report further provides key auxiliary indicators, including soil salinity accumulation, nitrate nitrogen leaching, irrigation water use efficiency, ecological risk standard deviation, and corresponding decision preference parameter values. Based on this, end users not only know what drip irrigation mode has been recommended, but also understand the expected yield, ecological cost, and stability level corresponding to the recommendation results, which facilitates subsequent field verification and management implementation.

[0212] 4) Develop a complete comprehensive evaluation report on the drip irrigation model. The comprehensive evaluation report should include at least a summary of the input scenario, the final optimized drip irrigation model parameters, the corresponding yield benefits, the corresponding ecological benefits, the main risk indicators, and the reasons for recommendation, so as to guide the actual control and promotion of the drip irrigation system.

[0213] like Figure 3 As shown, the Pareto optimal frontier is plotted with yield benefit (x-axis, kg / ha) as the x-axis and ecological benefit (y-axis, dimensionless) as the y-axis, plotting all candidate solutions (grey scattering points) and the Pareto optimal frontier (red line). The solutions on the Pareto frontier are all non-dominated solutions that do not dominate each other, demonstrating that multi-objective optimization can provide a set of balanced candidate drip irrigation modes for subsequent comprehensive evaluation and selection.

[0214] like Figure 4 As shown, the five key parameters of the final recommended drip irrigation mode are: irrigation volume (mm), irrigation interval (days), nitrogen fertilizer (kg / ha), phosphorus fertilizer (kg / ha), and potassium fertilizer (kg / ha). The orange area represents the agronomically feasible range, and the green bar represents the recommended value. The specific values ​​of the optimization results in the parameter space all fall within a reasonable range.

[0215] In a specific implementation plan, the evaluation and optimization of the drip irrigation model are as follows:

[0216] Based on the dynamic collaborative feature matrix constructed by the aforementioned steps, the particle swarm optimizer that integrates ecological model constraints, and the multi-objective evaluation system, the candidate optimized drip irrigation modes are verified, screened, and dynamically controlled in actual drip irrigation systems, realizing the effective transformation from theoretical optimization results to actual field operation.

[0217] In the specific implementation process, the final optimized drip irrigation model parameter vector determined after comprehensive scoring is first used as the initial recommended scheme. A small-scale verification test is carried out in the target plot or management zone. The verification test executes drip irrigation operations according to the irrigation volume, irrigation interval, and nitrogen, phosphorus, and potassium fertilizer application rates set by the optimized model. Crop growth parameters, ecological indicators, and final yield are monitored simultaneously throughout the crop's entire growth period. The measured results are compared and analyzed with the expected yield benefits and expected ecological benefits simulated by the optimized model in the model evaluation module to confirm whether the deviation between the simulated prediction and the measured results is within an acceptable range. If the deviation exceeds the preset threshold, the soil salinity balance model, crop growth model, and nitrogen balance model in the model evaluation module are locally parameterized using the measured data. The corrected model is then reconnected to the particle swarm optimizer, and the dynamic collaborative feature matrix is ​​updated using the current verification data as new samples. Local search and evaluation are performed again to obtain the corrected optimized model parameters, ensuring the adaptability of the recommended scheme to actual field conditions.

[0218] After verification and calibration, the final optimized drip irrigation mode is integrated into the intelligent control platform of the drip irrigation system. The control platform generates irrigation and fertilization instructions according to the optimized mode parameters, which are automatically executed by the irrigation control equipment. To achieve adaptive regulation in response to dynamic environmental changes, this invention further introduces a rolling optimization mechanism. Multiple decision nodes are set throughout the crop's entire growth period. At each decision node, the actual irrigation parameters, crop growth parameters, and ecological indicators collected up to the current moment are used as new samples. After standardization and dynamic time-series alignment, these are input into the dynamic collaborative feature construction process along with historical data to update the dynamic collaborative feature matrix. Based on this, using the current crop and ecological states as initial conditions, the particle swarm optimizer is invoked again to perform short-cycle rolling optimization of the drip irrigation mode parameters for subsequent stages. The newly generated optimization results are used as the control instructions for the next stage. During the rolling optimization process, the optimizer fully utilizes the inhibitory effect of ecological correction terms on the risk of soil salinity accumulation, ensuring that subsequent irrigation and fertilization operations maintain yield without leading to a continuous increase in ecological risk.

[0219] Furthermore, for application scenarios, this invention supports adaptive adjustment to different decision preferences. By modifying the risk preference factor and ecological risk aversion coefficient in the comprehensive scoring function, it outputs a customized drip irrigation mode recommendation set for different regions, soil types, or management objectives, and forms a list of regionally differentiated drip irrigation schemes for agricultural management departments or farmers to select according to their actual needs.

[0220] Ultimately, this invention outputs a complete report on the evaluation and optimization of drip irrigation models. The report includes at least the final optimized drip irrigation model parameters, corresponding predicted yield and ecological benefits, key ecological risk indicators, comparisons of verification test results, parameter adjustment records at each decision node during the rolling optimization process, and an evaluation of the recommended scheme's effectiveness in practical application. Through this process, the dynamic collaborative features, optimizer, and evaluation system constructed in the preceding steps are linked with the actual drip irrigation system in the field, enabling the optimization results to be implemented in a real-world environment and continuously improved based on monitoring data. This achieves a complete application of the drip irrigation model, from offline optimization to online adaptive control.

[0221] Example 2

[0222] A drip irrigation pattern evaluation and optimization system integrating an ecological model includes a module for executing processing instructions for each step in the drip irrigation pattern evaluation and optimization method integrating an ecological model.

[0223] The data acquisition and preprocessing module is used to collect multi-source data from irrigated plots, complete data labeling, standardization processing and outlier correction, and output a standardized multi-source data matrix.

[0224] The dynamic collaborative feature construction module is used to perform dynamic temporal alignment on standardized data, construct an adaptive kernel function with soil salinity constraints, and generate a dynamic collaborative feature matrix through kernel principal component analysis;

[0225] The particle swarm optimization module, which incorporates ecological constraints, is used to perform iterative optimization by introducing ecological model constraints, adaptive inertia weights, and ecological correction terms, and outputs a Pareto optimal solution set.

[0226] The multi-objective comprehensive evaluation module is used to construct a comprehensive evaluation system that includes yield, ecological benefits, and uncertainty penalties to screen the optimal drip irrigation mode;

[0227] The closed-loop control and output module is used for field verification and correction, intelligent platform integration and rolling optimization, and outputs customized solutions and application reports.

[0228] Although the specific embodiments of the invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the invention. Based on the technical solutions of the invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the invention.

Claims

1. A method for evaluating and optimizing drip irrigation patterns by integrating ecological models, characterized in that, include: Collect multi-source data from irrigated plots and perform standardized processing; The standardized data is dynamically aligned temporally and an adaptive kernel function is constructed to build a dynamic collaborative feature matrix. Constructing a dynamic collaborative feature matrix: The collected multi-source raw data are divided into raw irrigation parameter matrix, raw crop growth parameter matrix and raw ecological index matrix according to data type. Each matrix is ​​standardized by column, and the standardization results are checked for outlier consistency. The standardized irrigation parameter matrix, standardized crop growth parameter matrix and standardized ecological index matrix are output. Based on the growth patterns of target crops on irrigated plots, the entire growth period is divided into a unified stage sequence, resulting in a unified growth stage time axis. Based on this unified growth stage time axis, standardized multi-source data is reorganized into sample-level time-series vectors. Dynamic time warping is then performed on multiple types of time-series vectors within the same sample. Following the alignment path obtained from the dynamic time warping, the original unequal-length sequences are mapped onto the unified growth stage time axis. Soil salinity time-series data is extracted from this mapping to construct a soil salinity constraint sequence, where soil salinity data represents ecological indicator parameters from the multi-source data. Sample-level time-series vectors mapped onto the unified growth stage time axis are defined to have equal lengths and a one-to-one correspondence between stage points. The soil salinity constraint sequence is introduced, and an adaptive kernel function is constructed using the aligned sample-level time-series vectors as input. The aligned sample-level time-series vectors are concatenated in a fixed order. For any two samples, the kernel value between the two samples is calculated using an adaptive kernel function to obtain an adaptive kernel matrix. The adaptive kernel matrix is ​​then centered and eigenvalue decomposition is performed to obtain the kernel principal component directions sorted by contribution. The first few principal components are retained according to a preset rule. Each sample is projected onto the retained kernel principal component directions to obtain a dynamic collaborative feature matrix. Introducing ecological model constraints, adaptive inertia weights, and ecological correction terms into the particle swarm optimization framework, a particle swarm optimizer integrating ecological model constraints is constructed. Combining the dynamic collaborative feature matrix and crop growth model simulation results, the search direction of the particle swarm optimizer is normally corrected. The process of building a particle optimizer is as follows: The particle swarm optimizer is initialized, and the drip irrigation mode parameter vector is defined as the encoding method of the particles. The corresponding state vector in the dynamic collaborative feature matrix and the drip irrigation mode parameter vector of each particle are input into the model evaluation module, which includes the soil salinity balance model, crop growth model and nitrogen balance model. Based on the initial evaluation results of the model evaluation module, the individual historical optimal position of each particle is set, and an external non-dominated solution archive is established. Calculate the base velocity term and the ecological correction term, add the two together to get the new particle velocity, and set an upper limit for the velocity in each dimension; The particle position is iteratively updated based on the updated particle velocity. The feasible region is corrected for components that exceed the upper and lower limits of the parameters. A combination check is performed on the agronomic constraints that characterize the relationship between irrigation and fertilizer ratio. If the new position does not meet the combination constraints, it is projected to the feasible solution closest to the current point. The model evaluation module is called again. The historical best position of the individual particle is updated according to the Pareto dominance relationship. At the same time, the external archive is updated, the dominated solutions are deleted, and the new non-dominated solutions are retained. Repeatedly execute particle velocity update, particle position update, and archive update, and after multiple iterations, complete the optimization of the particle swarm; After particle swarm optimization, Pareto optimal solution is obtained, a comprehensive evaluation system is constructed, and the optimal drip irrigation mode is selected from the Pareto optimal solution; The optimal drip irrigation mode is evaluated and optimized. After field verification and correction, the optimal drip irrigation mode is integrated into the intelligent control platform. A rolling optimization mechanism is introduced for dynamic adaptive regulation and supports customized solution output. Finally, a drip irrigation mode evaluation and optimization application report is generated.

2. The method for evaluating and optimizing drip irrigation patterns using an integrated ecological model as described in claim 1, characterized in that, The calculation process for the basic velocity term and the ecological correction term is as follows: Based on the external non-dominated solution archive, a global guidance position is selected for each particle. The inertial weight is dynamically corrected based on the soil salinity component corresponding to the global guidance position, and the adaptive inertial weight is calculated. The basic velocity term is calculated based on the standard particle swarm update mechanism. This term includes the inertial inheritance term of the previous generation velocity determined by the adaptive inertial weight, the individual learning term toward the individual's historical best position, and the group learning term toward the global guiding position. The ecological costs at each reproductive stage are obtained through forward simulation. The ecological sensitivity vector is calculated by finite difference perturbation of decision variables. The correction term along the direction of ecological risk reduction is generated by combining the yield status adjustment correction intensity.

3. The method for evaluating and optimizing drip irrigation patterns using an integrated ecological model as described in claim 1, characterized in that, The process of obtaining the Pareto optimal solution is as follows: After particle swarm optimization is completed, all non-dominated solutions are extracted from the external non-dominated solution archive to form a Pareto optimal solution set. The Pareto optimal solution set is then deduplicated and merged with nearest neighbors. Pareto optimal solution set includes If two candidate drip irrigation modes have different parameters in each dimension, and the difference between them is less than the set tolerance, then the relatively stable candidate drip irrigation mode is retained.

4. The drip irrigation pattern evaluation and optimization method based on the integrated ecological model according to claim 1, characterized in that it comprehensively... The evaluation process is as follows: Based on the forward simulation results of each candidate drip irrigation mode using crop growth model, soil salinity balance model and nitrogen balance model, the yield benefits and ecological benefits of each candidate drip irrigation mode in the Pareto optimal solution set are calculated. The yield and ecological benefits of all candidate drip irrigation models in the Pareto optimal solution set were statistically analyzed to obtain the minimum, maximum, minimum, and maximum yield benefits, respectively. Uncertainty analysis was performed on each candidate drip irrigation model to obtain the standard deviation of ecological risk. For each candidate drip irrigation mode, a nonlinear utility mapping is performed on the yield benefit index to obtain the yield utility value; for each candidate drip irrigation mode, the ecological benefit index is normalized to obtain the ecological utility value. The yield utility value, ecological utility value, and ecological risk penalty are combined into a comprehensive score. The candidate drip irrigation mode with the highest comprehensive score is selected as the final recommended scheme from all candidate drip irrigation modes.

5. The method for evaluating and optimizing drip irrigation patterns using an integrated ecological model according to claim 4, characterized in that, The calculation process for yield benefits and ecological benefits is as follows: Pareto optimal solution set includes For each candidate drip irrigation mode, the crop growth model, soil salinity balance model, and nitrogen balance model are called for forward simulation. At the same time, the parameter vector of the candidate drip irrigation mode and the dynamic collaborative feature vector of the corresponding scenario are input. Based on the crop growth model, the yield benefit index of the candidate drip irrigation mode is output. The output of the soil salinity balance model is as follows: Soil salt accumulation for each candidate drip irrigation pattern, output according to the nitrogen balance model. The nitrate nitrogen leaching loss of each candidate drip irrigation mode is calculated, and the irrigation water use efficiency of the candidate drip irrigation mode is calculated based on the total yield and total irrigation water volume. Soil salinity accumulation and nitrate nitrogen leaching loss are converted into attenuation terms, and irrigation water use efficiency is converted into a saturation-type reward term. The three parts are combined to obtain the ecological benefit index. Finally, the yield and ecological benefit indicators for each candidate drip irrigation mode were obtained.

6. The method for evaluating and optimizing drip irrigation patterns using an integrated ecological model according to claim 1, characterized in that, The detailed report on the evaluation and optimization of drip irrigation models is as follows: Specific control parameters are read from the optimal drip irrigation mode, including the amount of water applied per irrigation, the irrigation interval, the amount of nitrogen fertilizer applied, the amount of phosphorus fertilizer applied, and the amount of potassium fertilizer applied. The report outputs the expected yield and ecological benefits under the final optimized drip irrigation model, as well as key auxiliary indicators, including soil salinity accumulation, nitrate nitrogen leaching, irrigation water use efficiency, ecological risk standard deviation, and corresponding decision preference parameter values, forming a complete comprehensive evaluation report of the drip irrigation model.

7. The method for evaluating and optimizing drip irrigation patterns using an integrated ecological model according to claim 1, characterized in that, The data collection process is as follows: Collect multi-source raw data covering multiple irrigation cycles or the entire growth stage, and systematically label the collected data samples; The multi-source raw data includes irrigation parameters from irrigation control equipment in the drip irrigation system, crop growth parameters and ecological indicators from field monitoring equipment; irrigation parameters include irrigation volume per irrigation, irrigation interval, nitrogen fertilizer application rate, phosphorus fertilizer application rate, and potassium fertilizer application rate; crop growth parameters include crop plant height, stem diameter, leaf area index, aboveground biomass, leaf chlorophyll content, and canopy temperature; ecological indicators include soil salinity, soil nitrate nitrogen content, soil moisture content, and groundwater depth; all observation records for each plot within a complete irrigation cycle or a complete growth stage are defined as a sample. Each sample was labeled with irrigation mode, crop response, and ecological feedback tags. A unified time index is established for the samples, and the entire monitoring period is divided into multiple consecutive reproductive stages or standardized time points. The collected multi-source data are associated with the time index according to their actual observation time points to form a labeled dataset.

8. A drip irrigation pattern evaluation and optimization system integrating an ecological model, comprising the drip irrigation pattern evaluation and optimization method integrating an ecological model as described in any one of claims 1-7, characterized in that, Includes the following modules connected in sequence: The data acquisition and preprocessing module is used to collect multi-source data from irrigated plots, complete data labeling, standardization processing and outlier correction, and output a standardized multi-source data matrix. The dynamic collaborative feature construction module is used to perform dynamic temporal alignment on standardized data, construct an adaptive kernel function with soil salinity constraints, and generate a dynamic collaborative feature matrix through kernel principal component analysis; The particle swarm optimization module, which incorporates ecological constraints, is used to perform iterative optimization by introducing ecological model constraints, adaptive inertia weights, and ecological correction terms, and outputs a Pareto optimal solution set. The multi-objective comprehensive evaluation module is used to construct a comprehensive evaluation system that includes yield, ecological benefits, and uncertainty penalties to screen the optimal drip irrigation mode; The closed-loop control and output module is used for field verification and correction, intelligent platform integration and rolling optimization, and outputs customized solutions and application reports.

Images